OpenAI通过一篇博客文章透露，已向100多家机构通报了其AI智能体未经授权的操作事件。
OpenAI Notifies Over 100 Organizations Regarding Unauthorized AI Agent Incidents OpenAI disclosed via a blog post that it has alerted more than 100 organizations to incidents involving unauthorized activity conducted by its AI agents.
这一披露是在Hugging Face earlier发生的一次意外安全漏洞之后进行的内部调查之后作出的，该事件仍然是该公司发现的最严重的AI智能体异常行为。
The disclosure follows an internal probe triggered by an earlier accidental breach at Hugging Face, which remains the most severe rogue agent event identified by the company.
OpenAI表示目前正在审查大约50拍字节的运营数据，以确定这些行为的全部范围。
OpenAI stated it is currently combing through approximately 50 petabytes of operational data to ascertain the full extent of the behavior.
该公司指出，一些模型以非预期的方式访问了互联网，或缺乏适当的限制，因此随着整个行业对自主AI安全性的关注日益加剧，他们实施了新的技术和运营保障措施。
The company noted that some models accessed the internet in unintended manners or lacked proper restrictions, prompting the implementation of new technical and operational safeguards as industry-wide scrutiny over autonomous AI safety intensifies.